WAS 7.0, SLES 10, Xen, and a petroglyph of a laptop

New toys to play with...

So, I managed to find an older T41 lappy that was doing a whole lot of nothing. I also found a DVD of SLES 10. With a bit of spare time on my hands, I now have a lab. Yay me.

Why do you care? Well, this is the second post on this blog, so you probably don't. But, I needed a sounding board to get some of the ticklers out of my brain. There's too much noise up there.

I have a theory that some of the overhead we see with WebSphere (and BEA, IIS, and any other highly threaded workloads) on VMWare will be dramatically less if we use the Xen hypervisor instead. It has to do with what I call 'the scheduled scheduler' problem. Basically, Xen's paravirtualization approach uses an integrated global scheduler for all of the guests. (It's not quite that simple, but this is an adequate explanation for now.) This is basically similar to pSeries LPARs in some respects, and AIX 6.1 WPARs in other respects.

On the other hand, VMWare ESX schedules guests based on various WLM weightings, and the guests in turn have their own schedulers. Like a small gear turning a big gear, it takes longer for the scheduled scheduler to schedule each of the workload threads.

So, my goal is to build out a benchmark environment on this T41. It has enough memory to be functional, but the CPU is nothing to write home about. I'm going to load WAS into the base OS as the control group, and WAS into a single Xen guest as the variable. I'll compare the performance of the control versus the variable groups and see what shakes out. Should be interesting, if you're sick like me.

In other news, WAS ND 7.0 is now in beta. Dare I use it for the Xen benchmark? Meh, we'll see.